In coding a moving picture, compression of information volume is usually performed by utilizing redundancies both in spatial and temporal directions which the moving picture has. Usually, a transformation into a frequency domain is used as a method of utilizing the spatial redundancy while inter picture prediction coding is used as a method of utilizing the temporal redundancy.
In a moving picture coding method which is presently under the process of standardization, quantization is performed on each block sized 4×4 pixels so as to generate coefficients after frequency transformation is performed on such block, with the view to enhance coding efficiency of a conventional MPEG-4 moving picture coding method. Then, scanning is performed starting at direct current components toward high frequency components, and combinations of a value R (Run, to be simply referred to as “R” hereinafter) indicating the number of consecutive zero coefficients and a coefficient value L (Level, to be simply referred to as “L” hereinafter) subsequent to it are generated so that a combination sequence (R, L) is made. After transforming this (R, L) into a code number using a predetermined code table, coding is performed by transforming the code number into a VLC code, further using a single Variable Length Coding (VLC) table. In the code table, a smaller code number is usually assigned as an occurrence probability gets higher. For example, a small code number is assigned to a combination where both R and L indicate small values since its occurrence probability is high. In certain VLC code tables, a VLC code having a short code length is assigned to a small code number (see reference to ISO/IEC 14496-2: “Information technology—Coding of audio-visual objects—Part2: Visual” 7.4.1, pp. 119-120, 1999.12).
However, using the existing method engenders a decrease in coding efficiency since the code length gets longer as the number of consecutive zero coefficients R and a coefficient value L get larger. Usually, the decrease in coding efficiency is obvious when a low frequency component value is coded since the coefficient value L as a low frequency component value is large.
Namely, as a result of assigning a single VLC table according to the occurrence probability and a single unique variable length code according to a pair of R and L, the coefficient value L indicating a large value is transformed into a variable length code having a very long code length. Even when coding L separately from R (one-dimensional coding of L) using a single VLC table, the same problem occurs as in the case of coding R and L as a pair.